Tasks / Experiment

Find the growth loop

Can the model find a product's real growth loop, show whether it compounds, and say which lever to pull?

Measures the modelTask type v1.0 · 2 tasksLast changed 2 Oct 2026 · ChangelogDifficulty

What AI gets right here, and what you’ll still have to catch

From 14 graded outputs by 7 models. 33% were usable with at most a quick edit.

Reliably right

  1. Sees the cross-side effect100% pass
    It traces the chain from the tutor bounty to oversupply, thinner bookings, new profiles without reviews, and weaker ranking, and acts on it by pausing broad tutor referrals.
    GPT-6.1 Sol · API · Growing on the surface, decaying underneath
  2. Addresses the actual decision96% pass
    The memo commits early to putting both engineers on idea 4, names the primary loop and its compounding status, and specifies what results would change the call (kill thresholds, quarter-end loop gain).
    Sonnet 5.5 · API · The badge on every form
  3. Produces the required deliverable96% pass
    The memo answers all parts of the brief (primary loop, compounding, engineer allocation, success measurement) in a usable form for the Head of Growth.
    Sonnet 5.5 · API · The badge on every form

Where it slips

  1. The loop maths holds46% pass
    The memo does not give a plain verdict of 'decaying' for the content loop despite showing its decline, and it does not compute a numeric yield or coefficient for that loop.
    GPT-6.1 Sol · API · Growing on the surface, decaying underneath
  2. Uses the supplied evidence correctly57% pass
    The claim that the base settles at 10,700 creators is unsupported by the pack's arithmetic, and the claim that cost per sign-up usually rises with spend is not in the supplied evidence.
    Opus 5.5 · Claude · The badge on every form
  3. Avoids unsupported claims59% pass
    Presents the 10,700 equilibrium and the rising cost-per-sign-up claim as facts without labelling them as hypotheses or supporting them from the pack.
    Opus 5.5 · Claude · The badge on every form

The tasks

Read the brief, then put up to three outputs side by side, each with the LLM judge’s verdict on every check. Highlights mark what a PM had to fix.

The brief

You're a Staff PM at Tutorly. Our CEO, Rachel Dunn, wants to triple paid acquisition for the next two quarters so we reach 50,000 booking parents before the Series B. Before the planning offsite she's asked you for an honest view of how we actually grow. Write a memo of no more than 1,300 words for Rachel and the exec team that: 1. Maps our growth loops in a simple text diagram, says which is primary and why, and whether each is compounding, contributing or decaying, with the numbers. 2. Responds to the plan to triple paid acquisition. 3. Says where our three squads should go for the next two quarters, with the test, threshold and stop condition for each. The pack is below. Not all of it matters equally.

What the model was given8 items: About Tutorly, Twelve months, at the top line, Search and profiles, Where new booking parents come from, and how long they stay, Referral programmes, Paid acquisition, Rachel's note, Constraints
About TutorlyAn online tutoring marketplace: parents book one-to-one lessons with tutors. The average lesson costs $40 and we keep a fixed 18% commission. Each tutor has a public profile page, with reviews from parents.
Twelve months, at the top lineBooking parents (at least one booking a month): 31,000 → 37,000 (+20%). Active tutors: 14,200 → 23,000 (+62%). Organic search sessions to tutor profiles: +40%. Bookings per active tutor a month: 9.1 → 6.8.
Search and profilesProfiles with three or more reviews get 81% of organic profile sessions; profiles with none get 4%. Of tutor profiles created in the last six months, 61% had no booking within 60 days, so they have no reviews. Click-through from search results on our top 200 keywords fell from 6.1% to 5.0% since March, after changes to the results pages.
Where new booking parents come from, and how long they stayOrganic search: 52% of new booking parents; 48% still booking after six months; net lifetime value $260. Parent referrals: 9%; 51% after six months; net lifetime value $275. Paid search and social: 39%; 22% after six months; net lifetime value $120. Across all sources, net lifetime value averages $210.
Referral programmesParents: $20 of lesson credit for each referred parent who books. Each active parent sends 0.31 invites a month, and 14% of invites become booking parents. Tutors: $50 for each referred tutor who completes onboarding. 44% of new tutors last year came through tutor referrals.
Paid acquisitionAverage cost per new booking parent over the year: $95. Last quarter we raised spend from $60,000 to $90,000 a month, and the cost per new booking parent rose from $82 to $109.
Rachel's note“Our LTV to CAC is 2.2. Every dollar we don't put into paid is growth we're leaving on the table. Triple it to $270k a month and we hit 50,000 parents before the raise.”
ConstraintsReviews can only be left after a completed, paid lesson. Our policy, and consumer protection rules in our main markets, forbid paying or rewarding anyone for reviews. The commission stays at 18%. Three squads are available for the next two quarters; the raise is about six months away.
What a strong answer doesThe answer key the graders mark against

Maps four loops: content (profiles and reviews rank, parents book, bookings create reviews, profiles rank better), parent referral, tutor referral, and paid. Names content as primary: organic is the biggest source of new booking parents (52%) and among the stickiest (48% at six months, $260). Finds that it's decaying behind 40% session growth. Profiles grew 62% against 40% sessions, so sessions per profile fell about 14%. 61% of new profiles never get a booking, so they never get reviews, and profiles without reviews get 4% of sessions. Bookings per tutor fell from 9.1 to 6.8. Traces the cross-side cause: tutor supply growing about three times faster than demand (+62% against +20%), much of it from the $50 tutor-referral programme (44% of new tutors), which spreads bookings too thin for new tutors to earn the reviews the loop runs on. Treats the click-through drop (6.1% to 5.0%) as a separate, external hit. Sizes the parent referral loop at about 0.043 (0.31 × 14%): small but high quality. Answers Rachel with marginal cost: the extra $30,000 a month bought about 94 more parents (60,000 ÷ 82 ≈ 732 → 90,000 ÷ 109 ≈ 826), roughly $319 each, against $120 of value for a paid parent. The 2.2 ratio blends sources and uses the average cost, so tripling would burn cash on parents who mostly leave. Sends the squads to: getting new tutors their first booking and review (for example, routing demand in under-supplied subjects to new tutors, with a threshold such as the share of new profiles booked within 60 days rising from 39% to 55%); concentrating search value on reviewed profiles (keeping unreviewed profiles out of the index, and a fix for the click-through drop); and pausing or targeting the tutor-referral bounty while improving the parent loop. Each comes with a stop condition, and the memo is honest about what 50,000 parents by the raise would really take, without rewarding reviews.

Critical failures (cap the score at 40)
  • Proposes paying, crediting or otherwise rewarding parents or tutors for reviews
  • Treats the 40% growth in organic sessions as proof the content loop is healthy
Task

Staff-level · v1.0 · synthetic · marketplace, network effects, SEO, unit economics, staff-level

The outputs

7 setups answered this brief. Pick up to three to read side by side.

Usable with a quick edit9 of 12 checks passed

The memo is well-structured and actionable; adding the per-profile session decline and an organic loop coefficient would be a quick edit without redoing the reasoning.

Output

To: Rachel Dunn and executive team From: Staff PM, Tutorly Subject: Grow completed lessons and repeat demand—not acquisition spend Decision requested: Do not triple paid acquisition. Allocate the three squads to organic conversion, marketplace activation, and parent referrals for the next two quarters.

1. How Tutorly actually grows

Our primary growth mechanism is successful lessons generating review-rich profiles that attract more parents. Paid acquisition feeds this mechanism; it is not itself a compounding loop.

```text PRIMARY: ORGANIC / REPUTATION Organic discovery → parent books → completed paid lesson → parent review ↑ ↓ └──────── stronger public profile / discovery ──────┘ ↓ repeat paid lessons

PARENT REFERRAL Booking parent → invitation → new booking parent → further invitations └──────────────── repeat lessons ──────────────────┘

TUTOR SUPPLY Tutor → tutor referral → onboarding → more available tutors ↓ [only valuable if matched to demand] ↓ lessons and tutor earnings

PAID INPUT Spend → new booking parent → lessons, reviews, repeat bookings and referrals ↑ ↓ └──────────────── reinvestable contribution ─────────────────────────┘ ```

Organic/reputation: primary, compounding in structure—but constrained

Organic supplies 52% of new booking parents, with 48% still booking at six months and $260 net LTV. Profiles with at least three reviews receive 81% of organic profile sessions; reviewless profiles receive only 4%. Successful transactions can therefore create an asset that attracts subsequent transactions.

However, this is evidence consistent with the loop—not proof that reviews alone cause traffic. Established tutors may differ in other ways.

The loop has two leaks:

  • 61% of profiles created in the last six months had no booking within 60 days. Without a completed paid lesson, they cannot earn reviews.
  • Search CTR on our top 200 keywords fell from 6.1% to 5.0%, an 18% relative decline. Restoring it would mean 22% more clicks at unchanged impressions, not necessarily 22% more bookings.

Organic sessions grew 40%, but we lack channel-specific booking-conversion history to establish whether the loop’s yield improved. This is our strongest compounding mechanism, not an unlimited growth engine.

Parent referrals: contributing, not demonstrated to be self-sustaining

Each active parent generates 0.31 × 14% = 0.0434 new booking parents per month. At 37,000 active parents, that is approximately 1,606 gross acquisitions monthly, assuming those rates hold.

A parent active for all six months would generate only 0.26 direct recruits in that period. We cannot calculate lifetime reproduction without active-lifetime data; current evidence does not establish a self-sustaining viral loop.

Nevertheless, referrals provide 9% of new booking parents, 51% six-month retention, and $275 net LTV—our best observed acquisition quality. The $20 credit is an incentive cost, not fully loaded CAC.

Tutor referrals: contributing supply, with decaying productivity

Tutor referrals generated 44% of new tutors, at $50 per completed onboarding. But onboarding does not create parent demand.

Active tutors rose 62%, versus 20% growth in booking parents. Monthly bookings per active tutor fell 25%, from 9.1 to 6.8. Multiplying the supplied figures implies total monthly bookings rose only about 21%. We are spreading demand across substantially more supply.

At $7.20 commission per lesson, monthly platform commission per active tutor fell from approximately $66 to $49, before costs. This is declining supply productivity, not proof that tutor-referral reproduction itself is shrinking. Broad supply recruitment should stop being a growth objective.

Paid: contributing acquisitions, decaying marginal efficiency

Paid contributes 39% of new booking parents, but only 22% remain booking after six months, with $120 net LTV. Its reinvestment loop has very little demonstrated surplus at current acquisition costs.

2. Why we should not triple paid

The quoted 2.2 LTV:CAC is $210 blended LTV divided by $95 paid CAC. It mixes populations. For paid parents, the historical ratio is $120/$95 = 1.26; at the latest CAC, it is $120/$109 = 1.10.

The recent spend increase is more concerning:

  • $60,000 at $82 CAC bought approximately 732 parents/month.
  • $90,000 at $109 bought approximately 826.
  • The additional $30,000 bought only 94 additional parents: approximately $320 marginal CAC.

That is an observational comparison, potentially affected by seasonality or mix—not a controlled estimate. It nevertheless argues against extrapolating the average CAC into a tripling.

Even assuming CAC stays at $109, $270,000 buys approximately 2,477 gross new parents/month. The increase over current spend buys about 9,900 additional parents over six months, before attrition. That does not independently close the 13,000 monthly-active-parent gap. Organic growth and reactivation may help, but acquisition totals are not active-parent totals.

Recommendation: Cap paid at no more than $90,000 monthly and remove unprofitable marginal campaigns now. Marketing, Finance and Analytics should run incrementality tests and build a source-specific cohort bridge: retained existing parents + retained acquisitions + reactivations = monthly booking parents.

For further scaling, propose a 1.5× net-LTV/incremental-CAC hurdle—CAC no greater than $80 at today’s paid LTV—with cash-payback visibility. This is a proposed risk buffer, not an observed benchmark. Earn increases in small steps; do not fund them against blended LTV.

3. Three squads, two quarters

The thresholds below are proposed decision rules. Pre-register cohort definitions, sample sizes and retention guardrails. Quarter one establishes causality; quarter two scales only passing treatments. Do not wait six months for every decision: use 60-day repeat booking as an interim guardrail, then validate mature cohorts.

Squad 1 — Organic discovery to first booking

Test: Diagnose the CTR decline, distinguishing search-engine changes from our own changes. Use matched query/page cohorts to test truthful snippets, relevant profile information and booking-path improvements. Do not assume we control external search-result layouts.

Threshold: At least 15% more first-booking parents per eligible organic impression versus control, without worse 60-day repeat booking. CTR recovery toward 6.1% is diagnostic, not the success metric.

Stop: End unsuccessful variants after a powered test, capped at 12 weeks; stop earlier for material booking or quality harm. In quarter two, expand winners across qualified profiles.

Squad 2 — First-lesson liquidity, not tutor headcount

Test: In demand-constrained categories/geographies, match suitable new tutors to genuine parent needs using availability, fit and booking support. Pause broad tutor-onboarding bounties; retain exceptions only for demonstrated supply gaps. Request honest reviews neutrally after completed paid lessons—never reward reviews.

Threshold: Reduce new-profile 60-day no-booking rate from 61% to 45% or below, while increasing total completed lessons per exposed parent by at least 10% and maintaining cancellation and repeat-booking quality.

Stop: Stop if activation merely redistributes existing bookings without increasing total lessons, or requires negative-contribution subsidies. Evaluate the first matured cohort in quarter one; scale only incremental wins in quarter two.

Squad 3 — Parent referrals after successful lessons

Test: Randomize referral prompts following successful lessons, simplify sharing and redemption, and retain the existing $20 booking-triggered credit.

Threshold: Increase invite-to-book yield from 0.0434 to at least 0.065 per active parent-month, with positive incremental contribution after credits and no deterioration in 60-day repeat booking.

Stop: Stop by week 12 if lift is attribution switching rather than incremental acquisition, economics fail, or fraud increases. Expand successful treatments in quarter two and validate six-month retention.

Bottom line: Our scarce resource is durable parent demand. Strengthen the transaction-to-reputation loop, improve matching, and amplify satisfied parents before buying substantially more low-retention traffic.

Grades and run details

Decision model 83 · LLM judge 10 of 13 checks
Decision model checks
  • passUses the supplied evidence correctly14%
  • passAddresses the actual decision97%
  • passRespects explicit constraints48%
  • passIdentifies material uncertainty94%
  • passAvoids unsupported claims42%
  • passProduces the required deliverable62%
  • partialFinds the decay behind the growth45%
  • passSees the cross-side effect28%
  • passAnswers the CEO on paid100%
  • partialA closed loop, not a channel64%
  • partialThe loop maths holds45%
  • partialProposes tests that could fail42%
Run
Run
#1
Time to output
71 s
Submitted
2 Oct 2026

Needs more than a quick edit10 of 12 checks passed

The memo contains an unsupported claim (14,000+ unreviewed profiles) and misstates the marginal CAC as $109, requiring correction before a PM could rely on it without re-checking the evidence.

Output

MEMORANDUM

TO: Rachel Dunn, CEO; Executive Team FROM: Staff Product Manager DATE: October 24, 2023 SUBJECT: Growth Engine Analysis, Paid Acquisition Response, and H1 Squad Allocations

*

Executive Summary

Tripling paid acquisition to $270k/month will not get us to 50,000 sustained booking parents for Series B. It relies on a metric illusion: our blended LTV is $210, but Paid LTV is only $120, while our marginal Paid CAC has already climbed to $109. Tripling spend will push marginal CAC well above LTV, burning cash to acquire low-intent parents who churn at more than double the rate of organic parents (22% vs. 48% 6-month retention).

Meanwhile, our true growth engine—the organic SEO/Review loop—is choking on a tutor supply glut that we are actively subsidizing. We must reallocate our squads to repair organic search distribution and tutor activation, turning our primary compounding loop back on before going out for Series B.

---

1. Tutorly’s Growth Engine: Loop Mapping & Health

``` [Loop A: SEO / Review Loop (Primary Engine)] Parent books lesson ──> Completes paid lesson ──> Leaves review ▲ │ │ ▼ Organic search session ◄── Profile ranks higher / gains CTR

[Loop B: Parent Referral Loop (Secondary)] Active parent ──> Shares invite ($20 credit) ──> New parent books

[Loop C: Paid Acquisition (Linear/Diminishing)] Ad spend ($) ──> Paid clicks ──> First-time booking parent (High churn)

[Counter-Loop: Supply Glut / Dilution] $50 Tutor referral bounty ──> +62% Tutors ──> 61% zero-booking profiles │ │ ▼ ▼ Search dilution & drop in tutor earnings ◄── Suppressed profile liquidity ```

#### Loop Health & Classification

LoopClassificationCurrent State & Metrics
:---:---:---
A. SEO / Profile Reviews (Primary)Decaying Compounding52% of new parents; $260 LTV; 48% M6 retention. Profiles with $\ge$3 reviews capture 81% of organic traffic, but 61% of new tutors get zero bookings in 60 days (0 reviews), starving the loop. Additionally, SERP CTR dropped from 6.1% to 5.0% post-redesign.
B. Parent ReferralsContributing9% of new parents; $275 LTV; 51% M6 retention. Each parent generates $0.31 \times 0.14 = 0.043$ new parents/month. Highly healthy unit economics, but lacks viral velocity ($K \ll 1.0$) to compound on its own.
C. Paid AcquisitionDecaying Linear39% of new parents; $120 LTV; 22% M6 retention. Marginal CAC rose from $82 to $109 when scaling to $90k/mo. Lacks compounding feedback; functions as a leaky bucket.
D. Tutor Supply BountyValue-Destroying44% of new tutors; $50 bounty. Supply surged +62% while demand grew +20%, depressing monthly bookings per tutor from 9.1 to 6.8 (-25%) and creating 14,000+ unreviewed "ghost" profiles.

Why Loop A is Primary: Organic search generates 52% of our demand at superior economics ($260 LTV vs. $120 Paid). It is our only structurally compounding loop: bookings generate reviews, which improve search ranking and page conversion, generating more bookings. However, it is currently decaying due to supply misallocation and SERP degradation.

---

2. Evaluating the Plan to Triple Paid Acquisition

Rachel's proposal states: "Our LTV to CAC is 2.2... Triple it to $270k a month and we hit 50,000 parents."

This relies on two critical errors:

1. The Blended LTV Fallacy: Rachel calculates LTV:CAC using blended LTV ($210) against historical blended CAC ($95) ($210 / $95 = 2.21$). But channel economics reveal a different reality: * Paid LTV is $120, driven by dismal 6-month retention (22%). * When we increased monthly spend from $60,000 to $90,000, our average Paid CAC rose from $82 to $109. * The marginal cost of those incremental acquisitions was: $$\text{Incremental CAC} = \frac{\$90,000 - \$60,000}{\Delta \text{Acquisitions}} = \frac{\$30,000}{\left(\frac{90,000}{109}\right) - \left(\frac{60,000}{82}\right)} = \frac{\$30,000}{825 - 731} \approx \$319$$ * At $109 average CAC against $120 LTV, current paid spend is barely breaking even ($1.1\times$). At $270k/month, Paid CAC will exceed Paid LTV on a fully loaded basis. 2. Quality of Growth for Series B: Pushing paid acquisition will inflate 30-day top-line numbers with unprofitable users who churn before the Series B diligence team reviews our cohorts. Series B investors will immediately spot cohorts decaying to 22% retention and discount our valuation.

Recommendation: Cap paid acquisition at $60,000/month (where CAC was sustainable at $82) and reallocate capital toward fixing marketplace liquidity and discovery.

---

3. Squad Allocation for the Next Two Quarters

To reach 50,000 booking parents sustainably, we must fix our core SEO loop and activate idle supply. We allocate our three squads as follows:

``` ┌─────────────────────────────────────────────────────────────┐ │ Squad 1: Marketplace Liquidity & Cold Start │ │ Objective: Solve the 61% unbooked tutor barrier │ ├─────────────────────────────────────────────────────────────┤ │ Squad 2: Search Experience & Organic Distribution │ │ Objective: Reverse SERP CTR drop from 5.0% to >6.5% │ ├─────────────────────────────────────────────────────────────┤ │ Squad 3: Parent Retention & Referral Compounding │ │ Objective: Expand organic retention and boost referral K-factor│ └─────────────────────────────────────────────────────────────┘ ```

#### Squad 1: Marketplace Liquidity & Cold Start * Problem: 61% of new tutors get no bookings in 60 days, yielding 0 reviews. Unreviewed tutors capture only 4% of traffic. We are paying $50 to onboard tutors into dead ends, depressing utilization to 6.8 bookings/tutor. * Focus: Route high-intent parent demand to unreviewed, vetted tutors using new-tutor introductory match discounts (funded by retiring the $50 tutor referral bounty) and smart onboarding guarantees. * Test: Launch a "First Lesson Booking Guarantee" matching algorithm on category pages, pairing new vetted tutors with price-sensitive parents. * Threshold: $\ge$35% of newly onboarded tutors secure their first completed lesson within 14 days (up from <15%), achieving $\ge$1 review within 21 days. * Stop Condition: If introductory match initiatives do not lift first-booking rates within 60 days without reducing existing tutor earnings below 6.5 bookings/month, halt automated discounting and shift to a curated intake cap.

#### Squad 2: Search Experience & Organic Distribution * Problem: Organic search profile sessions grew +40%, but CTR on top 200 keywords plummeted from 6.1% to 5.0% (-18% relative) following recent SERP changes. This represents hundreds of lost bookings per week. * Focus: Audit and roll back damaging SERP redesigns; implement rich review schema markups; optimize localized long-tail tutor directory pages (e.g., "Math Tutors in [City]"). * Test: A/B test SERP layout rollback vs. a structured review-first layout displaying verified subject ratings and badge endorsements compliant with consumer review regulations. * Threshold: Restore top-200 keyword CTR to $\ge$6.2% within 8 weeks, yielding $\ge$15% incremental organic booking volume. * Stop Condition: If technical SERP changes fail to move CTR by $\ge$0.5 percentage points after 6 weeks of clean indexation, freeze front-end adjustments and redirect squad to on-site landing page conversion and booking-funnel friction.

#### Squad 3: Parent Retention & Referral Compounding * Problem: Paid parent retention is poor (22%), while referral parents yield $275 LTV and 51% retention. However, referral volume is constrained (only 0.31 invites/parent/month). * Focus: Embed referral prompts directly after high-satisfaction milestones (e.g., 5-star lesson completions) and introduce parent-to-parent study group mechanics rather than generic link-sharing. * Test: Introduce dynamic milestone-based referral prompts (post-5-star rating) offering dual-sided $25 credits, combined with an automated re-booking reminder cadence for churn-risk cohorts. * Threshold: Increase invite velocity from 0.31 to 0.50 invites/active parent/month, sustaining an invite-to-booking rate $\ge$14% (lifting referral share of new parents from 9% to 15%). * Stop Condition: If CAC via referral incentives exceeds $45 per booking parent, or invite rate does not cross 0.40 within 60 days, kill the programmatic milestone incentives and pivot to lesson package re-engagement.

---

Series B Target Trajectory

MetricCurrentRachel's Paid PlanProposed Plan (6 Mo)
:---:---:---:---
Monthly Paid Spend$90,000$270,000$60,000
Blended M6 Retention41%28% (Paid heavy)47%
Bookings / Active Tutor6.86.28.5
Active Booking Parents37,000~48,000 (Leaky)46,500 (Durable)
Series B NarrativeUnprofitable growthHigh burn, low LTVHigh margin, compounding engine

By fixing our cold-start liquidity and reversing the SERP CTR loss, we will approach 47,000 highly retained, profitable booking parents. This presents a bulletproof Series B narrative: expanding gross margins, compounding organic discovery, and capital efficiency.

Claims the judge couldn’t find in the brief

  • marginal Paid CAC has already climbed to $109
  • creating 14,000+ unreviewed 'ghost' profiles

Grades and run details

Decision model 79 · LLM judge 10 of 13 checks
Decision model checks
  • passUses the supplied evidence correctly3%
  • passAddresses the actual decision99%
  • partialRespects explicit constraints19%
  • passIdentifies material uncertainty24%
  • failAvoids unsupported claims27%
  • passProduces the required deliverable76%
  • passFinds the decay behind the growth46%
  • passSees the cross-side effect100%
  • passAnswers the CEO on paid100%
  • partialA closed loop, not a channel45%
  • partialThe loop maths holds20%
  • passProposes tests that could fail46%
Run
Run
#1
API response time
34 s
Submitted
2 Oct 2026

Check by check

Where the outputs differ comes first. Open a check to read the judge’s reasons. “Mixed” means the two graders disagreed.

Where they differ 5

Uses the supplied evidence correctlyRightMixed
GPT-6 Astra · ChatGPT

All statements about the current situation are taken directly from the brief or derived by correct arithmetic.

Gemini 3.8 Flash · API

The claim '14,000+ unreviewed ghost profiles' is invented and not supported by the supplied context, and 'marginal Paid CAC has already climbed to $109' misstates the evidence (the $109 is average, not marginal).

Avoids unsupported claimsRightWrong
GPT-6 Astra · ChatGPT

Labels hypotheses and avoids presenting interpretations as established fact.

Gemini 3.8 Flash · API

Presents '14,000+ unreviewed ghost profiles' as fact without support, and states 'marginal Paid CAC has already climbed to $109' as a confident claim when the evidence only gives average CAC.

Finds the decay behind the growthWrongRight
GPT-6 Astra · ChatGPT

Does not compute the decline in sessions per profile (~−14%) despite having the 40% session growth and 62% tutor growth.

Gemini 3.8 Flash · API

Shows sessions per profile decline (implied by +40% sessions vs +62% tutors), 61% unbooked new profiles, and falling bookings per tutor, concluding the primary loop is decaying.

A closed loop, not a channelWrongRight
GPT-6 Astra · ChatGPT

Names the organic loop as primary but does not compute a yield or coefficient with retention applied.

Gemini 3.8 Flash · API

Names the SEO/review loop as primary closed loop, grounded in organic being the largest source with highest retention, and sizes it with 52% share and $260 LTV.

The loop maths holdsWrongRight
GPT-6 Astra · ChatGPT

Only computes the parent referral yield; missing organic loop coefficient and paid loop yield, so not every loop gets a computed yield.

Gemini 3.8 Flash · API

Parent referral yield (0.043) is computed correctly from 0.31 invites and 14% conversion; each loop gets a plain verdict (decaying compounding, contributing, decaying linear).

All got right 7

Addresses the actual decisionRightRight
GPT-6 Astra · ChatGPT

Commits early to not tripling paid and allocates squads, with conditions for scaling paid later.

Gemini 3.8 Flash · API

Commits clearly to not tripling paid, caps spend at $60k, and allocates three squads with explicit stop conditions that would change the call.

Respects explicit constraintsRightRight
GPT-6 Astra · ChatGPT

Memo is under 1,300 words, respects no review rewards, keeps commission at 18%, and uses three squads.

Gemini 3.8 Flash · API

Memo is within 1,300 words, addresses Rachel and exec team, does not propose paying for reviews, and keeps commission at 18%.

Identifies material uncertaintyRightRight
GPT-6 Astra · ChatGPT

Identifies causality uncertainty, missing conversion data, observational nature of marginal cost, and sets resolution thresholds.

Gemini 3.8 Flash · API

Each squad proposal includes a stop condition that names what result would change the approach, effectively identifying material uncertainties.

Produces the required deliverableRightRight
GPT-6 Astra · ChatGPT

Complete memo in the requested format, within length, and actionable by the exec team.

Gemini 3.8 Flash · API

The memo is complete, in the requested form, within length, and usable by the exec team with light edits.

Sees the cross-side effectRightRight
GPT-6 Astra · ChatGPT

Connects tutor referral bounty to oversupply, falling bookings per tutor, unbooked profiles, and lack of reviews weakening the content loop.

Gemini 3.8 Flash · API

Traces the $50 tutor bounty to oversupply (+62% tutors vs +20% parents), to thinner bookings (9.1→6.8), to new profiles without reviews, and proposes retiring the bounty.

Answers the CEO on paidRightRight
GPT-6 Astra · ChatGPT

Computes marginal cost ~$320 vs $120 paid parent value, explains why the 2.2 blended ratio misleads, and firmly says not to triple.

Gemini 3.8 Flash · API

Computes marginal cost (~$319) vs paid LTV ($120), explains the 2.2 ratio blends sources and averages, and firmly recommends not tripling paid.

Proposes tests that could failRightRight
GPT-6 Astra · ChatGPT

Each squad has a numeric threshold, a measurement window (12 weeks, 60-day guardrail), and a stop condition with clear actions.

Gemini 3.8 Flash · API

Each squad has a numeric threshold, a measurement window (e.g., 60 days, 8 weeks), and a stop condition that triggers a specific pivot.

Results

Every setup we’ve tested on this task type, across all its tasks and repeats, graded on the current checklist. Provisional The checklist is still being calibrated against our PM.

#Model · HarnessTask scoreDecision modelLLM judgeRunsCritical failures
1Sonnet 5.5withAPI87.396.22None
2GPT-6.1 SolwithAPI89.287.82None
3GPT-6 AstrawithChatGPT82.684.32None
4Opus 5.5withClaude87.164.42None
5GPT-6 LunawithAPI73.764.12None
6Gemini 3.8 FlashwithAPI64.676.92None
7Gemini 3.5 Flash-LitewithGemini40.953.82None

About the task

The PM job

Working out what actually drives growth, and where to push.

Why it matters

Teams tune funnel steps while the loop that compounds goes unmeasured. Mistaking a channel for a loop can cost a year.

What good looks like

  • A closed loop: each cycle's output feeds the next
  • The primary loop, traced from where the best users come from
  • The loop sized: cycle time, conversion, amplification
  • Retention in the maths
  • One lever, with a test that could fail

Deliberately not measured

  • Building a full growth model in a spreadsheet
  • Channel-level media planning
Capability tested

Growth systems thinking

The failure we’re looking for

Calls a channel a loop, or a referral button a viral loop

Grading

Decision model and LLM judge, calibrated against a blind PM review